Edinburgh Gut Cell Atlas and Common Coordinate Framework

The 1D-, 2D- and 3D-Gut Models and the GCA Common Coordinate Framework

The Human Cell Atlas resource will deliver single cell transcriptome data spatially organised in terms of gross anatomy, tissue location and with images of cellular histology. This will enable the application of bioinformatics analysis, machine learning and data-mining revealing an atlas of cell types and sub-types and ultimately disease conditions. However, to obtain an understanding of specific conditions, pathologies and histopathological phenotypes with their spatial relationships and dependencies we need a spatial descriptive framework to enable integration and analysis in spatial terms. For this we are developing a conceptual coordinate model of the application domain (small and large intestine) for spatial annotation of data samples and to develop the spatial descriptors for data retrieval, analysis and visualisation. The model will encode biologically and clinically relevant locations and regions and when a full HCA atlas framework becomes available the biological conceptual model will be mapped to the real-space common coordinate framework. This mapping from biologically natural coordinates to real-space image coordinates has been demonstrated for the mouse embryo and the same techniques will apply here.

For this project we are developing an explicit coordinate framework for data annotation that can be visualised in the simple 1-D conceptual linear gut, a 2-D “anatomogram” showing the key structures, landmarks and cell-types used for the ontology-based annotation and a full 3-D model of the gut based on a CT/MRI image of a normal gut. Here we discuss the development and implementation of the 1-, 2- and 3-D models as well as techniques for cross-mapping of the CCF between the different model representations

Image
Gut Atlas Coordinate Framework

 Figure 1 summarizes the underlying concept of the Edinburgh common coordinate framework (CCF) for the Helmsley Gut Cell Atlas project. All parts of the illustrated framework have now been implemented and can be explored with the browser-based viewer shown below 

In consultation with surgeons and clinical pathologists we developed a 1-D CCF capturing lineal distance in the gut from anus through the terminal ileum. In 2-D and 3-D this is represented as the centreline path through the gut and for sample location this is most useful metric for spatial comparison. Spatial location away from this path is captured using standard ontology terms for the gut layers, sub-layers, tissue and cell types as provided by the ASCT+B tables from the HCA/HuBMAP standards groups. The 1D CCF has been mapped to the 2-D Anatomogram developed for the Single Cell Expression Atlas and to four 3-D canonical models of the gut including both the male and female gut models from HuBMAP providing spatial integration and interoperability.

Locations within models are specified using the proportionate distances along midline paths between anatomical landmarks. Similarly, data are mapped between models using proportionate distances between landmarks along midline paths.

Locations within models are specified using the proportionate distances along midline paths between anatomical landmarks. Similarly, data are mapped between models using proportionate distances between landmarks along midline paths.Locations off of midline paths (but within the domain of the colon or terminal ileum) may be mapped to the midline using a precomputed map of the nearest midline point, where the nearest evaluation is constrained to the appropriate sub-domain and the sub-domains are the portions of the colon or terminal ileum that are well defined anatomical regions.

Image
Views through the Helmsley GCA 3D models for the common coordinate framework showing mapping between models
Image
Helmsley GCA CCF viewer

We have developed a web-based model viewer which will be used to display locations and mapped data in the context of our 1D and other mapped models. Here we show the 1D model above, a 2D anatomogram, a 3D model with inflated colon derived from an anonymised patient CT image and the HuBMAP VHF model. Where image data are associated with a model then image sections orthogonal to the midline paths may be viewed at all locations along the paths. Current locations and regions of interest are common to all linked models in the viewer so that selecting these in one model displays the corresponding locations and regions of interest in the other linked models. The viewer uses Fabric.js  and THREE.js  JavaScript libraries for interactive visualisation of 2 and 3D models. Details of anatomical structures, their locations and other configuration data unique to each model are obtained via an MVC based RESTful service layer with database connectivity developed as part of the project.

The prototype model explorer interface is now available and a poster presenting this work at the Helmsley Trust GCA Annual Meeting 2012 can beviewed and downloaded usng the link: